On Oct. 16, 2024, S&P Global Ratings published a request for comment (RFC) pertaining to the application of our global residential mortgage-backed securities (RMBS) criteria to U.S. RMBS transactions issued in 2009 and later that are backed by, or reference, residential mortgages. The global RMBS criteria provide an overarching framework for analyzing the credit risk for residential mortgage pools globally, which individual jurisdictions subsequently apply, taking into account country-specific features as provided in an associated supplement.

We believe the proposed expansion of the scope of the global RMBS criteria to include the U.S. and the proposed updates to the assumptions applied to U.S. RMBS will enhance the consistency and comparability across cohorts within residential mortgage portfolios within the U.S. as well as across residential mortgage markets globally. We are publishing this article to answer questions market participants may have regarding the RFC.

QUESTIONS AND ANSWERS

What are the key points of S&P Global Ratings' proposed framework for analyzing credit and cash flow risk in U.S. RMBS?

Our proposed framework remains broadly the same as the current one. In our credit analysis of transactions issued in 2009 and later that reference or are backed by a pool of U.S. residential mortgages (for ease of reference, we refer to these as "RMBS" in the remainder of this article), we assess the proportion of assets in a pool that we expect will default when subjected to different degrees of economic stress commensurate with our rating scenarios, as outlined in our ratings definitions article.

Thus, when analyzing whether an RMBS tranche can be assigned a given rating, we begin by estimating the gross level of defaults the residential mortgage pool may experience under a rating-specific economic stress. First, for an archetypal pool (an idealized version of the average pools historically observed in a particular jurisdiction; see details in the RFC), we apply an archetypal default assumption under the 'AAA' and 'B' scenarios. However, for a given pool, we apply loan-level adjustments for variations from the archetypal characteristics related to loan, borrower, or property. For each loan, we then determine the anticipated loss given default (loss severity) based on varying rating-level assumptions, such as market value declines of the property value and timelines for liquidation. The product of the default assumption and loss severity represents the loss projection (referred to as loss coverage) we would assess as a percentage of the balances of the loans and corresponding pool. The pool-level loss coverages are also adjusted based on additional analytical considerations including--but not limited to--geographic concentration, third-party due diligence results, the representation and warranty framework, and our view of the quality of the mortgage origination process.

We then perform a structural analysis where we apply the above-mentioned loss-coverage levels for different rating scenarios in conjunction with various cash flow assumptions (such as prepayment speeds, default timing, and interest rate patterns) to determine whether a given tranche receives contractual interest and principal in whole per the summation of our credit and cash flow stresses. Chart 1 provides a high-level overview of our analytical approach.

Chart 1

Could you elaborate on the reasons for the proposed update to the U.S. RMBS methodology?

First, although the years since our prior criteria calibration (which was based on data through mid-year 2015) have been marked by a generally favorable macroeconomic environment and supportive housing fundamentals (notwithstanding the COVID-19 pandemic), the passage of time has nonetheless provided meaningful insights. An additional eight years of performance data on pre-2009 originated loans since our prior calibration provided greater transparency on the ultimate resolution of severely delinquent loans given the share that were still in the loss-mitigation phase at the time of the prior calibration. This helped provide a better assessment of defaults for the most stressed 2006-2007 vintage loans.

Moreover, sufficient time (10 years) has passed to observe the impact of the regulatory changes in 2014, which, in our view, appear to have fundamentally improved mortgage industry practices. The introduction of the Ability-To-Repay/Qualified Mortgage (ATR/QM) rule, which applies to covered loans originated with application date of Jan. 1, 2014, and later, has demonstrated that it has significantly reduced commercial practices that weakened loan quality, including curbing the origination of riskier loan products prevalent pre-crisis, such as teasers and negative amortization loans. The consistent and broad use of third-party due diligence in the private-label securitization market in the last decade corroborates the improvement in the quality of underwriting, as supported by the generally low defect rates we've observed in transactions.

Further, given the expansion of our global RMBS criteria since its inception in 2019 to incorporate over 25 countries, our goal is to align the U.S.--as the sole outlier--with other jurisdictions in scope to enhance consistency and comparability across our RMBS ratings globally. In the process, we applied the mortgage market assessment (MMA) framework under the global RMBS criteria to better articulate our view of the risks of the U.S. mortgage market relative to the other jurisdictions in scope of the global framework.

What testing was conducted to evaluate the reasonableness of the proposed updates?

To support the reasonableness of the proposal, we back-tested sample transactions issued in 2006 and 2007 to compare the projected losses under the proposed updates to the actual realized losses.

Under our ratings definitions, macroeconomic conditions indicators--such as a decline in GDP and an increase in unemployment--would associate the Great Financial Crisis (GFC) with a 'BBB' stress level. However, a sector can experience higher or lower stress than the general macroeconomic stress. The observed losses in the residential mortgage sector associated with the GFC were higher than we would expect for a prudently functioning mortgage and housing market under a 'BBB' level of macroeconomic stress. This view is further supported by the material shift in the residential mortgage market driven by the regulatory enhancements over the last 10+ years compared to the prevailing mortgage market conditions leading up the GFC. Given this, we consider it reasonable to conclude that the mortgage and housing sector during the GFC was associated with stress levels greater than the general macroeconomy ('BBB').

Against the above backdrop, when back-testing a sample of RMBS transactions (with varying collateral characteristics) issued in 2006 and 2007, we compared the observed losses to those projected under the proposed updates in our loan-level credit model. The following scatter plot (see chart 2) displays the modeled loss rate under the proposed update on the vertical axis and the actual loss rates on the horizontal axis for the sample transactions. If a point falls above the line, it indicates the model over-projected the loss rate, while those points falling below the line indicate observed rates were higher than those projected by the model.

While the chart below shows an 'A' level of stress for comparison, certain observed losses may be higher or lower than the projected losses. In more cases than not, however, the projected losses under our proposed assumptions for a 'A' stress level are generally higher than those observed for these transactions. Moreover, the dispersion of some plots above or below the line also reinforces that specific transactions may display loss behavior that could also differ from an assumed set level of ratings stress for the broader residential mortgage and housing sector. In particular, the projected losses in the chart below only include geographic concentration adjustments as it pertains to pool-level loss projection overlays, and does not include overlays such as mortgage operational assessment, third-party due diligence, and representation and warranties, which also influence transaction-specific losses.

Chart 2

How did S&P Global Ratings arrive at lower default anchors for the archetypal pool?

Based on the following considerations, we proposed lowering the 'AAA' anchor FF to 14% from the current 15%, and the 'B' anchor to 2.0% from 2.5%.

An updated default definition that reflects the effect of loss-mitigation efforts and accounts for the overestimation bias for defaults under the prior default definition.  As noted above, eight years of additional performance data since the last calibration provided a better assessment of defaults (including the most stressed 2006-2007 vintage-originated loans). Specifically, the prior calibration considered a loan as defaulted if it ever experienced a 90+ day delinquency, irrespective of the ultimate loan resolution. By contrast, the current calibration considers a loan defaulted if the last observed payment status was 90+ days delinquent (including foreclosure and REO). This also includes the small percentage of loans that are still outstanding and are currently 90+ days delinquent (including foreclosure and REO).

Our assessment of what is considered a default for the purposes of our updated calibration analysis for this proposed update incorporates the loss-mitigation efforts (such as loan modifications) employed by servicers and the resulting population of loans that resolved in a payoff instead of liquidation. In our view, loss mitigation by servicers in the residential mortgage market comes with significant regulatory requirements and protocols that shouldn't be disregarded when projecting default behavior in mortgage credit analysis. When comparing empirically observed defaults using terminal loan status as the default definition for our calibration, we observed lower default rates than the ones in the current criteria (2018 default definition using an instance of 90+ days delinquent including foreclosure and REO). This further reinforces the proposed lower 14% 'AAA' anchor FF as the empirical default rate for peak-vintage loans with archetypal-like attributes was lower under the revised default definition applied in the updated calibration.

The lower proposed 'AAA' anchor FF is still within the range of observed mortgage defaults during the Great Depression (12%–16%).  When considering the 14% 'AAA' default assumption proposed in the RFC (compared to the current 15% assumption), we refer within the existing criteria to observed default rates for loans that we view as having archetypal characteristics of 12%-16%, and the definition of an archetypal loan isn't changing.

Impact of regulatory enhancements under the ATR/QM rule.  As noted above, 10 years on from the introduction of the ATR/QM rule, the regulation has thus far largely eliminated risky loan products prevalent before the GFC, such as teasers and negative amortization loans. The third-party due diligence in the private-label securitization market in the last decade also supports the improvement in the quality of underwriting, as demonstrated by generally low defect rates in transactions. Compared to the prior calibration in 2018, we have a longer period of available information that indicates an improvement in the underwriting standards.

MMA framework under the global RMBS criteria.  Archetypal default anchors: The global RMBS criteria outline the archetypal default rate range under different rating scenarios. We applied the framework in the global RMBS criteria to determine a country-specific MMA score (ranging from very low to very high in six ranges). This is largely based on a country's Banking Industry Country Risk Assessment (BICRA), which is subject to modifications to account for specifics of the residential mortgage market in the given country. We then considered the depth and breadth of market data available, as well as the historical performance of the assets, to determine the specific archetypal default rates within the applicable range for the resulting MMA score for the U.S.

As per the global RMBS criteria, we assessed the MMA for the U.S. residential mortgage market as Intermediate risk, which provides a 'AAA' anchor range of 10%-15% and a minimum 'B' default rate of 1.25%. Given the factors outlined above, we're proposing a 14% 'AAA' default anchor compared to the current 15%. The MMA is segmented into six categories (see table 1).

Table 1

The proposed 2.0% 'B' archetypal default rate (versus the current 2.5%) is largely based on 2001-2002 empirical default rates for archetypal-like loans and our short- to medium-term view of residential mortgage and housing fundamentals. Moreover, home prices remain resilient despite a rapid rise in mortgage rates in recent years based on the supply/demand fundamentals. As per the global RMBS criteria (and our current criteria), the archetypal default assumption at the 'B' rating level may adjust over time to reflect changing economic conditions when we believe they may materially affect levels of credit risk in the mortgage market. For example, during the recent COVID-19 pandemic, we temporarily raised our 'B' archetypal default rate for the U.S. When the 'B' archetypal default rate is updated, intermediate foreclosure frequency assumptions ('B+' to 'AA+') are also updated through an interpolation based on the relevant 'B' and 'AAA' values.

Table 2

What updates are proposed to the FICO/combined loan-to-value (CLTV) matrix and income verification/documentation FF adjustment factors?

In large part, the proposed FICO/LTV updates are rooted in the revised default definition applied in the updated calibration, with a greater reliance on the regression results compared to the prior calibration (further substantiated by the continued demonstration of the QM/ATR rule). Income verification and DSCR adjustment factor updates recognize the additional time horizon that has elapsed since such products were introduced, the commonality of these types in the market today, and the coinciding application of ATR for alternative income documentation loans. Furthermore, the factors are based on the consideration of the rank-ordering of credit risk between income fully verified using traditional methods (such as W-2s) and the comparative default ratios for very weakly underwritten income products pre-GFC.

Background--dataset and regression approach.  We performed a logistic regression analysis on approximately 10 million non-agency mortgage loans primarily originated from 2000-2008. The results of this analysis informed the updates to many of the FF adjustment factors. We focused on the loans that were originated pre-GFC, given the preponderance of data observations, degree of mortgage stress, and home price declines in the latter part of this origination vintage range. This analysis assisted in isolating the impact of different loan variables on defaults to provide additional insights into the incremental effect of non-archetypal loan attributes on performance. As noted above, we used a revised default definition (loans with the last observed payment status of 90+ days delinquent) in the updated regression analysis and performed additional analysis to develop the updated assumptions.

FICO/CLTV FF adjustment factors.  Our updated FICO/CLTV factors would generally be lower than those in the current criteria (see table 1 of the RFC for select FICO/CLTV combinations comparison). There are more material decreases for higher LTVs (above the neutral point of 75%) and lower FICO scores. Our calibration was informed by the above-mentioned logistic regression results. Given the limited data for higher LTVs (over 100%) regardless of FICO scores, for LTVs between 95% and 120%, the adjustment factors are derived by applying the growth rates of the factors from 70% LTV to 95% LTV linearly.

The table below illustrates the proposed impact on archetypal pool foreclosure frequency assumptions by varying FICO and CLTV attributes. The multiplicative effect of current FICO/LTV combinations using factors to the archetype also results in ceiling-like effects for which default projections reach 100%. For example, our current approach results in a multiplicative FF at 'AAA' of 123% (represented at the 100% cap in the table) for a 620 FICO and 100% LTV (8.2 * 15%).

Table 3a

Table 3b

Income verification/documentation foreclosure frequency adjustment factors.  We also updated the adjustment factors applied to post-GFC loan products, such as alternative income and DSCR loans. While we performed additional regression analysis that incorporated these assets, the limited degree of default observations, generally favorable economic conditions, and much smaller population than the pre-2009 origination vintages yielded less-informative results than that of the primarily 2000-2008 data set. However, we considered the pre-2009 regression results for weak income documentation loans in the historical data set for 2000-2008 vintages to develop an upper-end adjustment factor under the proposed criteria. We also considered that the "Other" doc type doesn't include adjustments for DTI and self-employed, and an alternative income loan would have DTI and self-employment factors. Based on these considerations, we're proposing to update the "Other" income verification type factor to 2.5x.

Therefore, the calibration includes a gradation of adjustment factors from neutral (1.0x) for full-income verification type (such as W-2/tax return) to 2.5x for "Other" income verification type, with alternative bank statement underwritten loans falling within the gradation range. While bank statement underwritten loans are considered to carry more credit risk than full-income verification methods (W-2, tax return, etc.), they have better characteristics than no-/stated income loans given actual income deposits, predominately sourced from third parties, and business expenses are used to derive a debt-to-income ratio.

In addition, for loans with full income verification, given an odds ratio close to neutral for loans with at least 12 but less than 24 months of income verification, we consolidated it with the 24+ months of full income verification category. For loans with less than 12 months of full income verification, we updated our adjustment factor to 1.2x from 1.5x. However, in recognition of the potential for greater volatility of income over time for non-salaried borrowers and considering the regression results, we also updated the self-employed adjustment factor, increasing it to 1.2x under the proposal from 1.1x.

For DSCR loans, we considered loans with a DSCR ratio of 0.70 and lower akin to the weakest income documentation category of 2.5x. For higher DSCRs, the DSCR factor would decrease but be no lower than 1.5x at a 1.3 and higher DSCR. In addition, the adjustment factor for investor occupancy (which is increasing to 2.0x from 1.5x) would also apply. By contrast, for the DSCR adjustment factors in the current criteria, the DSCR calibrated factor embeds investment occupancy status in the overall factor. The following chart provides the comparison of the factors (with the proposed DSCR factor being multiplied by the investment occupancy factor), which shows the maximum adjustment would be 5.0x, equating to the product of 2.5 (highest income factor) and 2.0 (investor occupancy factor).

Chart 3

Can you illustrate the application of the proposed U.S. RMBS methodology updates to different hypothetical mortgage pools?

The anticipated impact of the proposed updates would broadly indicate lower loss projections than the current ones. These changes stem primarily from the update to our archetypal default anchors for 'AAA' and 'B' (to 14% and 2%, respectively, from 15% and 2.5%), our updated FICO/LTV factors that--in most cases--provide for lower factors for FICO/LTV combinations (more noticeable at higher LTVs), and lower adjustment factors for certain product types (bank statement and DSCR). The following comparison shows the difference in loss projection estimates for four hypothetical pools under the current approach versus that in the RFC.

Table 4

Related Criteria

• Request For Comment: Global Methodology And Assumptions: Assessing Pools Of Residential Loans (U.S.), Oct. 16, 2024
• Global Methodology And Assumptions: Assessing Pools Of Residential Loans, Jan. 25, 2019